A method and apparatus for a k-way direct mapped cache organization is herein described. Control logic coupled to a cache may associate an address to a way within a plurality based on a first portion of the address. The control logic may match the first portion of the address to a predefined value in a mapping table, wherein the predefined value in the mapping table is associated with the way. In addition, the control logic may map the address to a set within cache based on a second portion of the address.
|
1. An apparatus comprising:
a microprocessor to generate a memory request referencing an address;
a memory array associated with the microprocessor logically organized into a plurality of sets and a plurality of ways; and
control logic to directly map a first portion of the address to a single way within the plurality of ways and a second portion of the address to a single set within the plurality of sets in response to the control logic receiving the memory request and by matching of the first portion to a cell of a mapping table of the control logic, the mapping table including a plurality of cells each to store a way value, wherein the mapping table is to associate a first part of the first portion of the address to a subset of the plurality ways, and to associate a second part of the first portion to a single way of the subset of ways, without a tag lookup.
12. An apparatus comprising:
a memory array comprising a plurality of elements, wherein the elements are addressable as a plurality of blocks, and wherein each element of the same offset within a block is addressable as a member of one of a plurality of sets; and
control logic to directly map a memory address to one of the plurality of blocks determined from a first number of bits in the memory address via matching of the first number of bits to a cell of a mapping table, the mapping table including a plurality of cells each to store a way value and to associate a first part of the first number of bits to a subset of the plurality ways, and to associate a second part of the first number of bits to a single way of the subset of ways, the control logic further to directly map one of the plurality of sets determined from a second number of bits in the memory address, and without a tag lookup.
21. A method comprising:
requesting an element from memory with a microprocessor, wherein the element is referenced by an address;
determining a single one of a plurality of ways within a cache associated with the microprocessor to check if the element is present therein based only on a first portion of the address, including reading the first portion of the address and matching the first portion of the address to a first value in a mapping table, wherein the mapping table associates the first value with one of the plurality of ways within the cache;
determining a single one of a plurality of sets within the cache to check if the element is present therein based only on a second portion of the address;
checking if the element is present only within the determined way and determined set in the cache without a tag lookup; and
providing the element to the microprocessor in response to the element being present within the determined way and set.
2. The apparatus of
3. The apparatus of
4. The apparatus of
5. The apparatus of
6. The apparatus of
7. The apparatus of
8. The apparatus of
10. The apparatus of
11. The apparatus of
14. The apparatus of
15. The apparatus of
16. The apparatus of
17. The apparatus of
18. The apparatus of
20. The apparatus of
22. The method of
reading the second portion of the address; and
matching the second portion of the address to a second value in the mapping table, wherein the mapping table associates the second value with one of the plurality of sets within the cache.
23. The method of
24. The method of
|
This invention relates to the field of cache memories and, in particular, to cache organization.
Providing design flexibility in caches by allowing a variety of size choices for the cache, while maintaining the speed of the cache in locating/storing a requested element, may be highly advantageous for architectures that utilize a cache. Traditionally, there have been three types of cache organizations that have been used: the fully associative, the k-way set associative; and the direct mapped cache organizations.
In a fully associative cache organization, each item of information from a main system memory is stored as a unique cache entry. There is usually no relationship between the location of the information in the cache and its original location in main system memory. Therefore, since each storage location can hold information from any location in main system memory, complex and expensive cache comparison logic is required to map the complete main system memory space. Furthermore, whenever a processor makes a memory request, each entry in a fully associative cache must be checked to see if the requested information is present (a hit), which forces a fully associative cache to stay extremely small as to not induce extremely large wait states in processing cycles.
The k-way set associative caches allow larger caches to be designed for a lower cost than fully associative caches, because less expensive and complex comparison logic is needed. Typically, a set associative cache divides the cache memory into k banks of memory, which is also known as k ways. To give a simplified example, if a 128 KB set associative cache has 4 ways, then each way may be 32 KB in size. Usually, a set associative cache sees memory as logically broken up into pages, which may be the size of each way. Continuing the example from above, a 256 KB main system memory may be logically viewed by the cache as 8 pages, each having a size of 32 KB.
Every location within a page of memory (such as the first location) may only be stored in the first location of each of the k ways. Therefore, in the example above, the first memory location in all 8 pages may be stored in only the first entry of any of the 4 ways. When a memory request is made, the set associative cache will compare the memory request with only the cache location the memory request would be stored at, in all of the ways. Since, the set associative cache need only compare the single location within each of the ways, the lookup times for memory requests are much quicker. These faster lookup times allow for larger set associative caches to be designed. However, the ability to compare locations in multiple ways still requires complex and expensive comparison logic.
One variation of the set associative cache that reduce the complexity and cost of the comparison logic is a direct mapped cache, which is effectively a one way set associative cache. Similar to the set associative cache, the direct mapped cache may view memory as broken into pages the size of the single way. From the example above, a 128 KB cache may have a single 128 KB way and may logically view a 256 KB memory as broken into two pages of 128 KB.
Yet, direct mapped caches may posses two major problems. First, when a program accesses two locations in memory, both at the same location in separate logically viewed pages, the direct mapped cache may have to update that location with every memory request (also known as thrashing). Thrashing eliminates any benefit of having a cache. Second, a direct mapped cache may be limited to a multiple of a power of two in design, since it contains only 1 way of 2s sets. As an example, if a processor has room on the die for a six megabyte (6 M) cache and a direct mapped organization is used, only a 4 M cache may be implemented. In contrast, a set associative cache organization may utilize all 6 M room for cache, because it divides the cache memory into k ways, each way having 2s sets (size of the cache=k*2s sets).
The present invention is illustrated by way of example and not intended to be limited by the figures of the accompanying drawings.
In the following description, numerous specific details are set forth such as examples of specific numbers of bits in an address, numbers of ways in a cache, placement of a cache, etc. in order to provide a thorough understanding of the present invention. It will be apparent, however, to one skilled in the art that these specific details need not be employed to practice the present invention. In other instances, well known components or methods have not been described in detail in order to avoid unnecessarily obscuring the present invention.
As stated above, microprocessor 105 may also have control logic 120 and cache 115. Although,
Cache 115 may be any memory, memory array, or any other storing structure commonly used for caching data or instructions. Referring to
An address 245 is also illustrated in
Turning to
Referring to
Location 430 may store any element, data element, instruction, operand, data operand, or any other item commonly stored in a memory location. Every memory location, such as location 430, may be referenced by an address, such as address 245. Control logic 120, shown in
As an illustrative example, if first predetermined portion 250 represents only page 410, and control logic 120 directly mapped the value of first predetermined portion 250 to way 205, then all of the locations within page 410 are placed in way 205. Therefore, if a request for an element within location 430 were issued, control logic 120 may only have to check way 205 to see if the element is present. As another example, if the value of first predetermined portion 250 references a plurality of pages and that value was directly associated with way 205, then all of the lookups for those locations and the storing of those locations may be limited to way 205. First predetermined portion 250 may also reference a certain offset within pages (a set), instead of a number of pages. An illustrative method for mapping a location to a way will be discussed in more detail in reference to
Turning to
In
Referring to
As noted above, a k-way direct mapped cache offers the ability to design a cache to any size (k*2s), such as in the set associative caches; however, the k-way direct mapped cache is able to avoid both the power of 2 limitation on traditional direct mapped caches and the expensive/complex lookup circuitry of a fully associative cache.
In the foregoing specification, the invention has been described with reference to specific exemplary embodiments thereof. It will, however, be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the invention as set forth in the appended claims. The specification and drawings are, accordingly, to be regarded in an illustrative sense rather than a restrictive sense.
Patent | Priority | Assignee | Title |
Patent | Priority | Assignee | Title |
5257361, | Oct 30 1989 | Hitachi, Ltd. | Method and apparatus for controlling one or more hierarchical memories using a virtual storage scheme and physical to virtual address translation |
5765199, | Jan 31 1994 | Motorola, Inc. | Data processor with alocate bit and method of operation |
5913222, | Apr 15 1994 | International Business Machines Corporation | Color correction method in a virtually addressed and physically indexed cache memory in the event of no cache hit |
6321297, | |||
6516386, | Dec 31 1997 | Intel Corporation | Method and apparatus for indexing a cache |
6751720, | Jun 10 2000 | Hewlett Packard Enterprise Development LP | Method and system for detecting and resolving virtual address synonyms in a two-level cache hierarchy |
20010026465, |
Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
Apr 16 2004 | DESAI, KIRAN R | Intel Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 015258 | /0356 | |
Apr 22 2004 | Intel Corporation | (assignment on the face of the patent) | / |
Date | Maintenance Fee Events |
Nov 13 2012 | ASPN: Payor Number Assigned. |
Nov 13 2012 | RMPN: Payer Number De-assigned. |
Jan 21 2015 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
Apr 01 2019 | REM: Maintenance Fee Reminder Mailed. |
Sep 16 2019 | EXP: Patent Expired for Failure to Pay Maintenance Fees. |
Date | Maintenance Schedule |
Aug 09 2014 | 4 years fee payment window open |
Feb 09 2015 | 6 months grace period start (w surcharge) |
Aug 09 2015 | patent expiry (for year 4) |
Aug 09 2017 | 2 years to revive unintentionally abandoned end. (for year 4) |
Aug 09 2018 | 8 years fee payment window open |
Feb 09 2019 | 6 months grace period start (w surcharge) |
Aug 09 2019 | patent expiry (for year 8) |
Aug 09 2021 | 2 years to revive unintentionally abandoned end. (for year 8) |
Aug 09 2022 | 12 years fee payment window open |
Feb 09 2023 | 6 months grace period start (w surcharge) |
Aug 09 2023 | patent expiry (for year 12) |
Aug 09 2025 | 2 years to revive unintentionally abandoned end. (for year 12) |